You bought the seats and ran the training. Turning either one into changed work takes a capability that sits with specific managers, and your plan did not test for it, name by name.
The Readout That Can’t Answer the Question
An executive walks into a quarterly review with a slide that holds up well. Licenses deployed. Training completed. Weekly active use climbing since spring. Somebody worked hard on that slide and the numbers on it are honest.
Then the question arrives, and it isn’t about adoption. Which workflow runs differently now than it did a year ago?
The room goes quiet in a specific way, and the reason has nothing to do with the AI underperforming. The honest answer is that people are producing the same work, in the same sequence, for the same consumers, with more help. The intake process has the same steps. The report goes out on the same day to the same distribution list. The review gate built around the old production step is still there, still staffed, still adding three days.
That gap between real tool adoption and unchanged work is what a seats-and-sessions plan buys. It doesn’t close by buying more of what produced it.
What a Seat and a Class Actually Buy
A license buys access. A prompt class buys tool familiarity. What they buy is a person who can operate the tool inside the job they already have, and for an individual contributor that’s a real investment: the analyst who drafts faster, the coordinator who summarizes a long thread accurately, the engineer who moves through boilerplate in a fraction of the time.
What neither one buys is a changed workflow. Nobody in that class was taught to take a process apart. Nobody was given the authority to end a report. Nobody was shown how to decide which steps a machine may now perform unsupervised and which ones a person still has to judge. Those are management acts, and they were never on the syllabus, because the syllabus was about the tool. A skills strategy that stops there has purchased capacity and left the question of what to do with it unassigned. Capacity that comes back from a tool gets absorbed by more of the same activity unless a leader decides otherwise. This article is about the layer that would have to make that decision, and whether it can.
The Capability Sits Lower Than the Budget
Approval for AI spending sits with the executive. Redesigning the work takes somebody close enough to it to know which steps exist and why, and positioned to change them.
That’s frequently a director, a team lead, a first-line supervisor, or a high-judgment individual contributor with a senior title and a small span of control. Anchor’s diagnostic for AI governance treats the accountable person as usually somebody inside the department rather than a chief, and often a fairly junior one, because that’s where the operating knowledge lives.
Which means a condition the return on an AI investment depends on sits below the person who signed for it, distributed across a population your own plan did not assess. It isn’t the only condition. A weak use case, data nobody can reach, or a control that doesn’t hold will defeat a capable manager, and Anchor’s framework carries five pillars for that reason. This is the one a seats-and-sessions plan leaves uncovered while appearing to address it.
Six Capabilities Worth Naming
The agent-management disciplines Anchor has published elsewhere describe what a function owes each AI agent it puts into production: an owner, a written standard, a review schedule, an escalation path, a retirement date. Those are artifacts a function establishes and maintains.
What follows is different. It’s what a manager has to be able to do, and the useful question about each one is behavioral: has the manager actually done it, not whether they agree it matters.
Redesign the work. Take a workflow apart, decide which steps merge, which move, which disappear, and which get a different owner, then run it in the new shape. The observable check is whether this manager has ever changed the sequence of a process rather than accelerating a step inside it. Task acceleration is what the tool sells. Sequence change is what the business feels.
Set the human judgment boundary. Decide where machine output is sufficient to act on and where a person still has to exercise judgment before anything leaves the building. Clark draws this line hardest at the licensed-professional edge, where AI reduces the preparation and the professional keeps the judgment, and the same discipline applies in any function with a consequential output. The observable check is where that boundary lives in the process, the review step, the sign-off, or the sampling rule, and the last time output was actually stopped there.
Evaluate the output. Know what good looks like for this specific work, well enough to write it down, and then sample real output against it on a schedule. A manager who can’t evaluate output can’t tell the difference between work that’s faster and work that’s better, and a weak performer using AI to skip the review step will look productive right up until somebody senior reads the result carefully. Enthusiasm for the tool reads like this capability and can be what conceals its absence.
Coach a team through a changed workflow. Hold people through the months where the old way still works and the new way isn’t comfortable yet. That means calibrating two different problems at once: the person who won’t trust the output and re-derives it by hand, and the person who trusts it completely and stopped reading it. Both are coaching problems with different conversations. The observable check is naming a person moved off one of those two positions, and what the manager actually did to move them.
Tell an owner apart from a coordinator. When the redesign gets assigned, hand it to somebody who controls what the change requires: the queue, the review rule, the staffing priority, the report’s existence. Accountability given to a person who has to ask somebody else for every lever is coordination wearing an owner’s title, and the work stalls in a third party’s inbox. The observable check is whether this manager, the last time they delegated something structural, gave the person the levers with it.
Retire the old work. End the report, the reconciliation step, the review gate, or the standing meeting that the change made unnecessary. This is the hardest of the six and it’s rarely a reasoning failure. Ending a report removes somebody’s visibility, and that somebody doesn’t always rank below the manager making the call. The observable check is the simplest one on the list: name the last thing this manager stopped.
Why the Gap Stays Invisible
Three reasons, and none of them involve anybody behaving badly.
Readiness answers lose an edge at every layer they pass through. Ask a manager whether their team is ready and the answer is shaped by what the manager believes is expected of them. Aggregate that up three levels and it arrives in the executive suite smoothed into confidence. Clark’s stated view is to treat any workforce-readiness figure that came from the top with skepticism: he’s describing what happens to a subjective judgment as it passes through people who all want the answer to be yes, not describing dishonesty.
Capability gets confused with fluency. The manager who’s excited about AI, who found the tool, who shows people tricks in the team meeting, reads as the capable one, and may be. Tool fluency and the ability to restructure work are separate skills, and they’re only loosely correlated.
Nobody has been asked, on the form that would count. Performance reviews measure the job as it was defined, with no line on whether this manager can redesign it, and no forum where anyone was ever required to answer that honestly, by name.
The same skepticism has to reach the top of the house. The six capabilities describe executive behavior inside a smaller span of control too, and the readiness-inflation mechanism above means the executive floor has not been checked any more carefully than the layer below it.
Capability and Authority Are Different Findings
A manager can hold all six capabilities and still change nothing.
If the queue belongs to another function, the review rule belongs to compliance, and the report belongs to a senior executive who likes receiving it, then the manager is blocked. That’s an authority finding, and the remedy is executive backing secured in advance, not development.
The distinction matters because these two findings look identical from the top. Both present as a workflow that didn’t change. Conflating them turns a capability review into an unfair performance review, which is the fastest way to make the whole exercise unusable, because the next round of honest answers won’t arrive.
So the diagnosis needs both halves. Can this manager do it, and can this manager reach the things it requires? A yes to the first and a no to the second is an executive problem. A no to the first is a development problem. A no to both is why the same workflow keeps coming back on the agenda without ever moving.
Run It as an Inventory
Run the six capabilities as an inventory against people, by name, rather than as a checklist for the organization.
Take the managers who own the workflows where your AI investment is supposed to show up first, however many people that turns out to be. For each one, mark each capability present, unproven, or absent, based on something they’ve actually done rather than something they’d say in an interview. Then mark whether they can reach the levers.
Those three marks are not interchangeable, and keeping them apart is what separates a planning exercise from an unfair review. Mark a capability present when you can point at the behavior: this manager, this work, this change. Evidence that is missing, out of your reach, or inconclusive stays unproven, which is where most of a first pass will land. The bar for absent is higher than a blank cell. It takes a fair, bounded attempt in which the manager had the relevant authority and the support to use it, and still could not do the thing. An empty record is not an absence finding.
The first honest pass will produce a lot of “unproven,” because most executives haven’t watched most of these managers attempt any of the six. That’s still a finding, and what it reports first is your own evidence rather than your managers. Across a whole function, unproven says you currently can’t see this capability being exercised. That may be because the redesign work never came up, because it happened under a different label, or because the record sits with a leader you didn’t ask. A missing redesign pipeline is one diagnosis worth testing there, not a conclusion the blank cells prove. Test it by going after operating history and by creating bounded chances to demonstrate the capability. One evidence source is already available: what changed in this manager’s area over the last two years, and who drove it. That answers more of the grid than an interview would.
Expect the result to come out uneven, and treat the unevenness as the finding. One manager will have five of six and no authority. Another will have redesigned processes for fifteen years and never once written down what good output looks like. A third will be strong on evaluation and unable to end anything. That pattern tells you where the next investment belongs far more precisely than an adoption chart does, and it’s available to you this quarter without a new tool, a new hire, or a new program.
Run this as planning, not as a personnel review, and tell the people being asked that’s what it is. None of it is a headcount argument. Five of the six capabilities describe the same team doing different or better work, and the sixth describes ending work the redesigned process made unnecessary.
When the Gap Is Real
Three management responses, none of which requires a new enterprise program.
Develop it in place. A supported redesign, not a training session, is how this capability gets built and demonstrated. Give the manager one bounded workflow, a defined window, and a senior reviewer chosen by the same demonstrated-behavior standard as the inventory itself. If nobody senior passes that standard, that’s the first gap to close, and the pairing move below goes up a level instead. Clark’s framing on executive judgment is that it’s muscle memory, exercised or not, and the same is true here. One supported redesign doesn’t complete the capability. It’s where it starts.
Pair it. Name the capability out loud, then run one bounded workflow with the experienced manager accountable and the developing manager holding the pen. When that workflow ships, accountability transfers to the developing manager for the next one, with the experienced manager reviewing rather than doing. The pairing ends when a redesign ships without the escort. The manager’s manager sets both workflows and calls the transfer, and nobody has to be reassigned for it.
Move the accountability. If the workflow matters this quarter and the manager who owns it can’t yet run the change, the change goes to somebody who can. This one usually needs an executive rather than the manager’s manager, because the levers it requires often sit in another function. The executive names the temporary redesign owner, loans that person the specific levers the change requires, the queue, the review rule, the staffing priority, the report’s existence, and states the condition that ends the loan, which is the redesigned workflow running in its new shape. Two accountabilities run at once here and they should be labeled that way. The seated manager keeps the operating result and the day-to-day number. The temporary owner keeps the redesign until the end condition is met, then hands the levers back. Announce the arrangement to the team, not just to the two managers. That’s a sequencing decision rather than a judgment on the person, and it should be said that way.
More training is worth doing once the capability question has an answer. On its own, it’s what produced the slide in the opening scene.
Before the Next Round of Licenses
Pull the last AI skills plan your organization approved and read it as a purchase order. Count what’s in it: seats, tiers, sessions, completion targets, maybe a certification. Then look for the line that names a manager and a capability.
If that line isn’t there, the plan bought access to a tool and left the work alone. That’s a planning gap, fixable in the next cycle, not a verdict on the people running it.
Anchor’s AI Bearing Assessment publishes Talent and Skills as one of its five pillars: “Who in the building can build with AI, who can govern it, and where the critical capability gaps sit.” That last clause is the one this article has been about.
Pick one workflow. Name the manager who owns it. Answer the two questions: can they do it, and can they reach what it requires.